A Monocular Vision-Based Human-Following Approach for Mobile Robots

The ability to recognize and follow human beings is considered to be a key technology for mobile robots. A vision-based following approach is proposed in this paper. It consists of two parts: the human detection part and the visual servoing part. In the human detection part, the image information is captured by the robot camera and processed by a neural network detection algorithm to extract human position and state information. In the visual servoing part, robot switches between three modes including seeking, following and stopping for human-following. Technically, an algorithm is proposed to calculate the angle between the robot and the human, which is used to control. The innovation of this paper is the use of a full vision-based approach to human-following, where the whole process only requires image information. A series of experiments were conducted to verify the effectiveness of the approach.